With the widespread application of unmanned aerial vehicles (UAVs) in various fields, anomaly detection in flight data has become increasingly important. However, temporal variations and temporal correlations in flight data challenging to achieve accurate anomaly detection. To address these issues, the multiscale Transformers with contrastive learning for UAV anomaly detection (MTCL-UAV) is proposed to achieve adaptive multiscale modeling. The model is based on the mixture-of-experts (MoEs) architecture, and each MoE block includes a router, expert networks (ENs), and an aggregator. The router performs a temporal decomposition to select optimal scales for each sample, and the aggregator fuses the outputs of the ENs into an integrated representation. To learn the temporal correlations of flight data and improve the model’s representation, a dual attention mechanism enhanced by contrastive learning (CL-DAM) is introduced, which captures not only intrapatch and interpatch correlation relationships but also neighborhood relationships among patches. Experiments on a real flight dataset demonstrate that MTCL-UAV not only achieves superior performance compared to other methods but also exhibits strong robustness. The code is available at https://github.com/SteelHu/MTCL-UAV

To address the issues of structural diversity, modeling complexity, and the lack of evaluation methods in drone swarm command and control (C2) networks, this paper proposes a complex network-based quality evaluation method for drone swarm C2 networks from a network topology perspective. First, by analyzing the structure of the drone swarm C2 system, three hierarchical C2 network models are constructed, which are based on the Leader–Follower architecture, BA scale-free network, and ER random network, respectively. Subsequently, a drone swarm network quality evaluation indicator, system integrating network connectivity, load status, and transmission efficiency is established, along with an evaluation model that considers both static and dynamic characteristics. Finally, an analysis is conducted using networks of the same scale but different C2 structures. The evaluation results demonstrate that this method can effectively distinguish the performance of networks with different structures and exhibits good applicability under both random and targeted attack scenarios. Under static scenarios, distributed C2 networks exhibit the highest quality values, while centralized networks demonstrate the lowest. In random attack scenarios, the Leader–Follower structure achieves the highest network quality among the three hierarchical architectures, outperforming BA and ER network structures by 117% and 25%. In targeted attack scenarios, the ER network structure achieves the highest network quality, surpassing Leader–Follower and BA network structures by 66% and 17%. It provides a quantitative reference for the design and optimisation of the drone swarm C2 system structure.

With the development of sensor and communication technologies, unmanned aerial vehicles (UAVs) have been widely applied in various critical fields. UAV anomaly detection (AD) with flight data can quickly identify system failures and prevent major accidents and losses. Existing methods typically ignore scale variations and intricate spatio-temporal dependencies inherent in-flight data, and they depend on fixed thresholds that often produce high false alarms. To address these issues, we propose a multiscale spatio-temporal graph convolutional network (MSTGCNet) for UAV AD. First, the graph-enhanced mixture-of-experts (GMoE) block uses seasonal-trend decomposition to route each input sequence to the expert network (EN) whose patch-based receptive field best matches its temporal scale, enabling effective multiscale feature extraction. Second, the causally constrained spatio-temporal GCN (CC-STGCN) jointly learns an adaptive spatio-temporal graph and applies graph convolutions to capture dependencies across sensors and time. Third, we devise an adaptive thresholding strategy for streaming data (ATSSD) that dynamically adjusts detection thresholds within a sliding window based on local statistics, reducing false positives. Experiments on three real-world UAV flight datasets demonstrate that MSTGCNet significantly outperforms state-of-the-art baselines in $F1$ score, showcasing its excellent overall performance, and the ablation study validates the effectiveness of the model’s key components. Finally, parameter sensitivity analysis and visualization experiments provide deep insights into the proposed model and the characteristics of the flight data. The code is available at https://github.com/SteelHu/MSTGCNet